datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
lang5_probes
Selected Probes
Each probe is a CSV with prompt, prompt_len, and target columns. All targets are 0/1 integers unless noted. All datasets are balanced (50/50) unless noted.
5 — hist_fig_ismale
Entries: 5,000 | Avg prompt length: 20 chars | Max: 70 chars
Prompts: Historical figure names (e.g. "Margaret of Clisson", "Billy Mays").
Target: 1 = male, 0 = female — 50% / 50%
6 — hist_fig_isamerican
Entries: 5,000 | Avg prompt length: 17 chars | Max: 65 chars… See the full description on the dataset page: https://huggingface.co/datasets/timaeus/lang5_probes.tsd2026-verbatim-leakage-probes
TSD 2026 Verbatim Leakage Probes
High-entropy prefix–suffix probes for measuring verbatim training-data leakage in
open-weight language models. This is the evaluation set used in Context as a Key:
Quantifying Verbatim Data Leakage Across Model Scale, Alignment, and Reasoning
Architectures (TSD 2026).
5,000 unique probes × 3 prefix lengths = 15,000 rows.
What this is for
Each row gives you a prefix to feed a model and a target_suffix the model never
saw. If the… See the full description on the dataset page: https://huggingface.co/datasets/fremy7/tsd2026-verbatim-leakage-probes.svo_probes
SVO-Probes
This dataset comes from https://github.com/deepmind/svo_probes.
Usage
from datasets import load_dataset
# Note that the following line says "train" split, but there are actually no splits in this dataset.
dataset = load_dataset("MichiganNLP/svo_probes", split="train")
# To see an example, access the first element of the dataset with `dataset[0]`.
chessgpt-board-probes-dataset
ChessGPT Board Probes Stockfish Dataset
This dataset contains fixed-length PGN strings used to train and test linear probes
on TransformerLens-converted chess models.
Train: stockfish_train.csv (10,000 games)
Test: stockfish_test.csv (10,000 games)
Each PGN string is exactly 365 characters long.
Source: internal subset derived from Stockfish-vs-Stockfish self-play games.
For usage examples, see the jd0g/chessgpt-board-probes model repo.
Uploaded at: 2025-08-29T03:58:31.828571Z
bangla-agreement-probes
Bangla agreement & honorific-register minimal pairs
4,790 hand-built minimal pairs probing Bangla subject-verb person/honorific
agreement (distance-binned: none/short/medium/long), agreement attraction,
and cross-sentence pro-drop register agreement. Native-speaker reviewed.
Built for a controlled Transformer vs Mamba-3 vs Hybrid case study.
File
Phenomenon
Pairs
sva.tsv
subject-verb agreement, distance-binned
3300
attraction.tsv
agreement attraction… See the full description on the dataset page: https://huggingface.co/datasets/sahilfarib/bangla-agreement-probes.datasets